機械知覚&ロボティクスグループ
中部大学

Deep Learning 国際会議

Larval Hostplant Prediction from Luehdorfia japonica Image using Multi-label ABN

Author
Tsubasa Hirakawa, Takaaki Arai, Takayoshi Yamashita, Hironobu Fujiyoshi, Yuichi Oba, Hiromichi Fukui, Masaya Yago
Publication
ECCV Workshop on Computer Vision for Ecology, 2024

Download: PDF (English)

Butterflies are easily recognizable due to their showy coloration, and are a familiar taxon to many enthusiasts. Because of the abundance of specimens collected and the ease of comparison among various species, regional variation in butterfly spots can be observed, which is related to multiple factors such as geological history, topography, climate, and hostplants. In this study, we focus on the relationship between regional variation in butterfly spots and the distribution of larval hostplants and aim to clarify the relationship by classifying the larval hostplants based on images of butterfly spots. Specifically, we focus on the Luehdorfia japonica, a species of butterfly with known geographic variation in butterfly spots and a highly understood distribution. We create Luehdorfia japonica image dataset based on digital specimens and the metadata about the collection site. We show that the multi-label attention branch network can be trained on the dataset to accurately classify the larval hostplant from the specimen images and that the analysis of the attention map provides the same basis for decision making as the expert knowledge.

前の研究 次の研究